The researchers at the ICAHN Medicine School have developed a powerful AI tool, based on the same transformer architecture used by large language models such as Chatgpt, to process a whole night of sleep. To date, it is one of the largest studies, which analyzes 1,011,192 hours of sleep. The details about their findings were reported in the March 13 edition of the magazine Sleep.
The model, called the fundamental transformer of the sleep patch (PFTSELP), analyzes brain waves, muscle activity, heart rate and breathing patterns to classify sleep stages more effectively than traditional methods, rationalize sleep analysis, reduce variability and support future clinical tools to detect sleep disorders and other health risks.
The current sleep analysis is often based on human experts who manually write down short sleep data segments or the use of AI models that are not able to analyze a patient’s sleep. This new approach, developed using thousands of sleep recordings, has a more complete vision. When training in full -length sleep data, the model can recognize sleep patterns throughout the night and in different populations and environments, offering a standardized and scalable method for sleep research and clinical use, researchers say.
“This is a step forward in the analysis and interpretation of the dream assisted by AI,” says the first author Benjamin Fox, a doctorate candidate at the ICAHN School of Medicine in Mount Sinai in the artificial intelligence training area and emerging technologies. “By taking advantage of AI in this way, we can learn clinical characteristics directly from the data of the sleep study and use them for sleep score and, in the future, other clinical applications such as detecting sleep apnea or evaluating health risks linked to sleep quality.”
The model was built using a large set of sleep studies data (polysomnograms) that measure key physiological signals, including brain activity, muscle tone, heart rate and breathing patterns. Unlike traditional AI models, which analyze only short 30 second segments, this new model considers all night of the sleep, capturing more detailed and nuanced patterns. In addition, the model is trained through a method known as self-supervision, which helps learn relevant clinical characteristics of physiological signals without using results labeled by humans.
“Our findings suggest that AI could transform the way we study and understand the dream,” says the corresponding co-senior author, Ankit Parekh, PhD, assistant medical professor (pulmonary, critical care and sleep medicine) in the ICAHN School of Medicine in Mount Sinai and director of the circadian and circadian analysis group in Mount Sinai. “Our next goal is to refine technology for clinical applications, such as the identification of sleep -related health risks more efficiently.”
The researchers emphasize that this AI tool, although promising, would not replace the clinical experience. Instead, it would serve as a powerful help for sleep specialists, helping to accelerate and standardize sleep analysis. Next, the team’s research aims to expand its abilities beyond the classification of the sleep stage to detect sleep disorders and predict health results.
“This AI promoted approach has the potential to revolutionize sleep research,” says the corresponding co-senior author Girish N. Nadkarni, MD, MPH, president of the Department of Artificial Intelligence and Human Health of Windreich at the Icahn School of Medicine, director of the Hasso Plattner Institute for Digital Health for Digital Health for Digital Health for Digital Health, and Irene and Dr. Arthur M. Medicine Professor Fisberg. Dr. Nadkarni is also the inaugural chief of the Digital Medicine and co -director -based data division and co -director of the Mount Sinai Clinical Intelligence Center. “When analyzing the entire nights of sleep with greater consistency, we can discover deeper ideas about sleep health and its connection with general well -being.”